Mathematical proofs in ArXiv papers recognized by OCR have appeared with irrelevant Chinese.

[bwnjnOEI]

Error screenshot

Code snipp

from pix2text import Pix2Text

img_fp = 'paper/2402.11867v3.pdf'
p2t = Pix2Text.from_config()
doc = p2t.recognize_pdf(img_fp)
doc.to_markdown(f'./{img_fp}.md')

Qeustion

Are Chinese and English models separate or a single model?
What parameters or models can be adjusted to achieve better results?
Will paid models definitely avoid this situation?

[breezedeus]

Initalize it with these configs:

import os
from pix2text import Pix2Text

text_formula_config = dict(
    languages=('en', ),  # 设置识别的语言
    text=dict(
        rec_model_name='en_PP-OCRv3',
        rec_model_backend='onnx',
    ),
)
total_config = {
    'layout': {'scores_thresh': 0.45},
    'text_formula': text_formula_config,
}
p2t = Pix2Text.from_config(total_configs=total_config)

[bwnjnOEI]

Initalize it with these configs:

import os
from pix2text import Pix2Text

text_formula_config = dict(
    languages=('en', ),  # 设置识别的语言
    text=dict(
        rec_model_name='en_PP-OCRv3',
        rec_model_backend='onnx',
    ),
)
total_config = {
    'layout': {'scores_thresh': 0.45},
    'text_formula': text_formula_config,
}
p2t = Pix2Text.from_config(total_configs=total_config)

I appreciate your feedback, I changed the setting according to your suggestion, but the situation still occurs:

[breezedeus]

Check the logs printed during model initialization to see if the model 'en_PP-OCRv3' is really being used. Judging from your result, it seems that this configuration is not effective. If it is not effective, you can try the following initialization code:

import os
from pix2text import Pix2Text

text_formula_config = dict(
    languages=('en', ),  
    text=dict(
        rec_model_name='en_PP-OCRv3',
        rec_model_backend='onnx',
        cand_alphabet=None,  # NOTE: must add this line
    ),
)
total_config = {
    'layout': {'scores_thresh': 0.45},
    'text_formula': text_formula_config,
}
p2t = Pix2Text.from_config(total_configs=total_config)